The report of the second campaign of the evaluations, kept

[PierreMardon]

The report of the second campaign of the evaluations, kept for the next one to be compared with: evals/reports/2026-10-03-e830bc236e9b/. It measures the chain after the four first fixes drawn from the first campaign (#78, #79, #80, #81), on 01-overdue-list and 03-overdue-reminders, three runs each: 67 sessions, 17.07 USD at list price.

What was done

• report.md and summary.json of the campaign, as report --against evals/reports/2026-10-03-6bf7c06609a6/summary.json --keep wrote them. Nothing else changes.

What it shows

• Every run reached conformant and passed all of its hidden acceptance tests.
• What moved since the first campaign, by the report's own rule: on 01-overdue-list the interview asks one question where it asked none, the order of ties, in three runs of three; on 03-overdue-reminders the cut of the blueprint and the messages score one point better.
• What did not: blueprint.no-padding is 2 on every run, the extractor being unchanged; following.no-noise is 2; the developer still sends one blueprint back per run on average.
• First spreads between runs of one case: a run of 01-overdue-list costs from 1.41 to 3.60 USD and takes from none to two corrections.

The conclusions, issue by issue, are in the issues #77 lists.

Tests

No code changes. The text rules of the repository pass on the two files; the CI runs every gate.

Refs #77